1
# Copyright (C) 2005, 2006, 2007 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
# 10:16 < lifeless> make partial index writes safe
56
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
57
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
59
# move sha1 out of the content so that join is faster at verifying parents
60
# record content length ?
63
from cStringIO import StringIO
64
from itertools import izip, chain
68
from bzrlib.lazy_import import lazy_import
69
lazy_import(globals(), """
89
from bzrlib.errors import (
97
RevisionAlreadyPresent,
99
from bzrlib.osutils import (
106
from bzrlib.versionedfile import (
107
AbsentContentFactory,
111
FulltextContentFactory,
117
# TODO: Split out code specific to this format into an associated object.
119
# TODO: Can we put in some kind of value to check that the index and data
120
# files belong together?
122
# TODO: accommodate binaries, perhaps by storing a byte count
124
# TODO: function to check whole file
126
# TODO: atomically append data, then measure backwards from the cursor
127
# position after writing to work out where it was located. we may need to
128
# bypass python file buffering.
130
DATA_SUFFIX = '.knit'
131
INDEX_SUFFIX = '.kndx'
134
class KnitAdapter(object):
135
"""Base class for knit record adaption."""
137
def __init__(self, basis_vf):
138
"""Create an adapter which accesses full texts from basis_vf.
140
:param basis_vf: A versioned file to access basis texts of deltas from.
141
May be None for adapters that do not need to access basis texts.
143
self._data = KnitVersionedFiles(None, None)
144
self._annotate_factory = KnitAnnotateFactory()
145
self._plain_factory = KnitPlainFactory()
146
self._basis_vf = basis_vf
149
class FTAnnotatedToUnannotated(KnitAdapter):
150
"""An adapter from FT annotated knits to unannotated ones."""
152
def get_bytes(self, factory, annotated_compressed_bytes):
154
self._data._parse_record_unchecked(annotated_compressed_bytes)
155
content = self._annotate_factory.parse_fulltext(contents, rec[1])
156
size, bytes = self._data._record_to_data((rec[1],), rec[3], content.text())
160
class DeltaAnnotatedToUnannotated(KnitAdapter):
161
"""An adapter for deltas from annotated to unannotated."""
163
def get_bytes(self, factory, annotated_compressed_bytes):
165
self._data._parse_record_unchecked(annotated_compressed_bytes)
166
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
168
contents = self._plain_factory.lower_line_delta(delta)
169
size, bytes = self._data._record_to_data((rec[1],), rec[3], contents)
173
class FTAnnotatedToFullText(KnitAdapter):
174
"""An adapter from FT annotated knits to unannotated ones."""
176
def get_bytes(self, factory, annotated_compressed_bytes):
178
self._data._parse_record_unchecked(annotated_compressed_bytes)
179
content, delta = self._annotate_factory.parse_record(factory.key[-1],
180
contents, factory._build_details, None)
181
return ''.join(content.text())
184
class DeltaAnnotatedToFullText(KnitAdapter):
185
"""An adapter for deltas from annotated to unannotated."""
187
def get_bytes(self, factory, annotated_compressed_bytes):
189
self._data._parse_record_unchecked(annotated_compressed_bytes)
190
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
192
compression_parent = factory.parents[0]
193
basis_entry = self._basis_vf.get_record_stream(
194
[compression_parent], 'unordered', True).next()
195
if basis_entry.storage_kind == 'absent':
196
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
197
basis_lines = split_lines(basis_entry.get_bytes_as('fulltext'))
198
# Manually apply the delta because we have one annotated content and
200
basis_content = PlainKnitContent(basis_lines, compression_parent)
201
basis_content.apply_delta(delta, rec[1])
202
basis_content._should_strip_eol = factory._build_details[1]
203
return ''.join(basis_content.text())
206
class FTPlainToFullText(KnitAdapter):
207
"""An adapter from FT plain knits to unannotated ones."""
209
def get_bytes(self, factory, compressed_bytes):
211
self._data._parse_record_unchecked(compressed_bytes)
212
content, delta = self._plain_factory.parse_record(factory.key[-1],
213
contents, factory._build_details, None)
214
return ''.join(content.text())
217
class DeltaPlainToFullText(KnitAdapter):
218
"""An adapter for deltas from annotated to unannotated."""
220
def get_bytes(self, factory, compressed_bytes):
222
self._data._parse_record_unchecked(compressed_bytes)
223
delta = self._plain_factory.parse_line_delta(contents, rec[1])
224
compression_parent = factory.parents[0]
225
# XXX: string splitting overhead.
226
basis_entry = self._basis_vf.get_record_stream(
227
[compression_parent], 'unordered', True).next()
228
if basis_entry.storage_kind == 'absent':
229
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
230
basis_lines = split_lines(basis_entry.get_bytes_as('fulltext'))
231
basis_content = PlainKnitContent(basis_lines, compression_parent)
232
# Manually apply the delta because we have one annotated content and
234
content, _ = self._plain_factory.parse_record(rec[1], contents,
235
factory._build_details, basis_content)
236
return ''.join(content.text())
239
class KnitContentFactory(ContentFactory):
240
"""Content factory for streaming from knits.
242
:seealso ContentFactory:
245
def __init__(self, key, parents, build_details, sha1, raw_record,
246
annotated, knit=None):
247
"""Create a KnitContentFactory for key.
250
:param parents: The parents.
251
:param build_details: The build details as returned from
253
:param sha1: The sha1 expected from the full text of this object.
254
:param raw_record: The bytes of the knit data from disk.
255
:param annotated: True if the raw data is annotated.
257
ContentFactory.__init__(self)
260
self.parents = parents
261
if build_details[0] == 'line-delta':
266
annotated_kind = 'annotated-'
269
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
270
self._raw_record = raw_record
271
self._build_details = build_details
274
def get_bytes_as(self, storage_kind):
275
if storage_kind == self.storage_kind:
276
return self._raw_record
277
if storage_kind == 'fulltext' and self._knit is not None:
278
return self._knit.get_text(self.key[0])
280
raise errors.UnavailableRepresentation(self.key, storage_kind,
284
class KnitContent(object):
285
"""Content of a knit version to which deltas can be applied.
287
This is always stored in memory as a list of lines with \n at the end,
288
plus a flag saying if the final ending is really there or not, because that
289
corresponds to the on-disk knit representation.
293
self._should_strip_eol = False
295
def apply_delta(self, delta, new_version_id):
296
"""Apply delta to this object to become new_version_id."""
297
raise NotImplementedError(self.apply_delta)
299
def line_delta_iter(self, new_lines):
300
"""Generate line-based delta from this content to new_lines."""
301
new_texts = new_lines.text()
302
old_texts = self.text()
303
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
304
for tag, i1, i2, j1, j2 in s.get_opcodes():
307
# ofrom, oto, length, data
308
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
310
def line_delta(self, new_lines):
311
return list(self.line_delta_iter(new_lines))
314
def get_line_delta_blocks(knit_delta, source, target):
315
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
316
target_len = len(target)
319
for s_begin, s_end, t_len, new_text in knit_delta:
320
true_n = s_begin - s_pos
323
# knit deltas do not provide reliable info about whether the
324
# last line of a file matches, due to eol handling.
325
if source[s_pos + n -1] != target[t_pos + n -1]:
328
yield s_pos, t_pos, n
329
t_pos += t_len + true_n
331
n = target_len - t_pos
333
if source[s_pos + n -1] != target[t_pos + n -1]:
336
yield s_pos, t_pos, n
337
yield s_pos + (target_len - t_pos), target_len, 0
340
class AnnotatedKnitContent(KnitContent):
341
"""Annotated content."""
343
def __init__(self, lines):
344
KnitContent.__init__(self)
348
"""Return a list of (origin, text) for each content line."""
349
lines = self._lines[:]
350
if self._should_strip_eol:
351
origin, last_line = lines[-1]
352
lines[-1] = (origin, last_line.rstrip('\n'))
355
def apply_delta(self, delta, new_version_id):
356
"""Apply delta to this object to become new_version_id."""
359
for start, end, count, delta_lines in delta:
360
lines[offset+start:offset+end] = delta_lines
361
offset = offset + (start - end) + count
365
lines = [text for origin, text in self._lines]
366
except ValueError, e:
367
# most commonly (only?) caused by the internal form of the knit
368
# missing annotation information because of a bug - see thread
370
raise KnitCorrupt(self,
371
"line in annotated knit missing annotation information: %s"
373
if self._should_strip_eol:
374
lines[-1] = lines[-1].rstrip('\n')
378
return AnnotatedKnitContent(self._lines[:])
381
class PlainKnitContent(KnitContent):
382
"""Unannotated content.
384
When annotate[_iter] is called on this content, the same version is reported
385
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
389
def __init__(self, lines, version_id):
390
KnitContent.__init__(self)
392
self._version_id = version_id
395
"""Return a list of (origin, text) for each content line."""
396
return [(self._version_id, line) for line in self._lines]
398
def apply_delta(self, delta, new_version_id):
399
"""Apply delta to this object to become new_version_id."""
402
for start, end, count, delta_lines in delta:
403
lines[offset+start:offset+end] = delta_lines
404
offset = offset + (start - end) + count
405
self._version_id = new_version_id
408
return PlainKnitContent(self._lines[:], self._version_id)
412
if self._should_strip_eol:
414
lines[-1] = lines[-1].rstrip('\n')
418
class _KnitFactory(object):
419
"""Base class for common Factory functions."""
421
def parse_record(self, version_id, record, record_details,
422
base_content, copy_base_content=True):
423
"""Parse a record into a full content object.
425
:param version_id: The official version id for this content
426
:param record: The data returned by read_records_iter()
427
:param record_details: Details about the record returned by
429
:param base_content: If get_build_details returns a compression_parent,
430
you must return a base_content here, else use None
431
:param copy_base_content: When building from the base_content, decide
432
you can either copy it and return a new object, or modify it in
434
:return: (content, delta) A Content object and possibly a line-delta,
437
method, noeol = record_details
438
if method == 'line-delta':
439
if copy_base_content:
440
content = base_content.copy()
442
content = base_content
443
delta = self.parse_line_delta(record, version_id)
444
content.apply_delta(delta, version_id)
446
content = self.parse_fulltext(record, version_id)
448
content._should_strip_eol = noeol
449
return (content, delta)
452
class KnitAnnotateFactory(_KnitFactory):
453
"""Factory for creating annotated Content objects."""
457
def make(self, lines, version_id):
458
num_lines = len(lines)
459
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
461
def parse_fulltext(self, content, version_id):
462
"""Convert fulltext to internal representation
464
fulltext content is of the format
465
revid(utf8) plaintext\n
466
internal representation is of the format:
469
# TODO: jam 20070209 The tests expect this to be returned as tuples,
470
# but the code itself doesn't really depend on that.
471
# Figure out a way to not require the overhead of turning the
472
# list back into tuples.
473
lines = [tuple(line.split(' ', 1)) for line in content]
474
return AnnotatedKnitContent(lines)
476
def parse_line_delta_iter(self, lines):
477
return iter(self.parse_line_delta(lines))
479
def parse_line_delta(self, lines, version_id, plain=False):
480
"""Convert a line based delta into internal representation.
482
line delta is in the form of:
483
intstart intend intcount
485
revid(utf8) newline\n
486
internal representation is
487
(start, end, count, [1..count tuples (revid, newline)])
489
:param plain: If True, the lines are returned as a plain
490
list without annotations, not as a list of (origin, content) tuples, i.e.
491
(start, end, count, [1..count newline])
498
def cache_and_return(line):
499
origin, text = line.split(' ', 1)
500
return cache.setdefault(origin, origin), text
502
# walk through the lines parsing.
503
# Note that the plain test is explicitly pulled out of the
504
# loop to minimise any performance impact
507
start, end, count = [int(n) for n in header.split(',')]
508
contents = [next().split(' ', 1)[1] for i in xrange(count)]
509
result.append((start, end, count, contents))
512
start, end, count = [int(n) for n in header.split(',')]
513
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
514
result.append((start, end, count, contents))
517
def get_fulltext_content(self, lines):
518
"""Extract just the content lines from a fulltext."""
519
return (line.split(' ', 1)[1] for line in lines)
521
def get_linedelta_content(self, lines):
522
"""Extract just the content from a line delta.
524
This doesn't return all of the extra information stored in a delta.
525
Only the actual content lines.
530
header = header.split(',')
531
count = int(header[2])
532
for i in xrange(count):
533
origin, text = next().split(' ', 1)
536
def lower_fulltext(self, content):
537
"""convert a fulltext content record into a serializable form.
539
see parse_fulltext which this inverts.
541
# TODO: jam 20070209 We only do the caching thing to make sure that
542
# the origin is a valid utf-8 line, eventually we could remove it
543
return ['%s %s' % (o, t) for o, t in content._lines]
545
def lower_line_delta(self, delta):
546
"""convert a delta into a serializable form.
548
See parse_line_delta which this inverts.
550
# TODO: jam 20070209 We only do the caching thing to make sure that
551
# the origin is a valid utf-8 line, eventually we could remove it
553
for start, end, c, lines in delta:
554
out.append('%d,%d,%d\n' % (start, end, c))
555
out.extend(origin + ' ' + text
556
for origin, text in lines)
559
def annotate(self, knit, key):
560
content = knit._get_content(key)
561
# adjust for the fact that serialised annotations are only key suffixes
563
if type(key) == tuple:
565
origins = content.annotate()
567
for origin, line in origins:
568
result.append((prefix + (origin,), line))
571
# XXX: This smells a bit. Why would key ever be a non-tuple here?
572
# Aren't keys defined to be tuples? -- spiv 20080618
573
return content.annotate()
576
class KnitPlainFactory(_KnitFactory):
577
"""Factory for creating plain Content objects."""
581
def make(self, lines, version_id):
582
return PlainKnitContent(lines, version_id)
584
def parse_fulltext(self, content, version_id):
585
"""This parses an unannotated fulltext.
587
Note that this is not a noop - the internal representation
588
has (versionid, line) - its just a constant versionid.
590
return self.make(content, version_id)
592
def parse_line_delta_iter(self, lines, version_id):
594
num_lines = len(lines)
595
while cur < num_lines:
598
start, end, c = [int(n) for n in header.split(',')]
599
yield start, end, c, lines[cur:cur+c]
602
def parse_line_delta(self, lines, version_id):
603
return list(self.parse_line_delta_iter(lines, version_id))
605
def get_fulltext_content(self, lines):
606
"""Extract just the content lines from a fulltext."""
609
def get_linedelta_content(self, lines):
610
"""Extract just the content from a line delta.
612
This doesn't return all of the extra information stored in a delta.
613
Only the actual content lines.
618
header = header.split(',')
619
count = int(header[2])
620
for i in xrange(count):
623
def lower_fulltext(self, content):
624
return content.text()
626
def lower_line_delta(self, delta):
628
for start, end, c, lines in delta:
629
out.append('%d,%d,%d\n' % (start, end, c))
633
def annotate(self, knit, key):
634
annotator = _KnitAnnotator(knit)
635
return annotator.annotate(key)
639
def make_file_factory(annotated, mapper):
640
"""Create a factory for creating a file based KnitVersionedFiles.
642
This is only functional enough to run interface tests, it doesn't try to
643
provide a full pack environment.
645
:param annotated: knit annotations are wanted.
646
:param mapper: The mapper from keys to paths.
648
def factory(transport):
649
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
650
access = _KnitKeyAccess(transport, mapper)
651
return KnitVersionedFiles(index, access, annotated=annotated)
655
def make_pack_factory(graph, delta, keylength):
656
"""Create a factory for creating a pack based VersionedFiles.
658
This is only functional enough to run interface tests, it doesn't try to
659
provide a full pack environment.
661
:param graph: Store a graph.
662
:param delta: Delta compress contents.
663
:param keylength: How long should keys be.
665
def factory(transport):
666
parents = graph or delta
672
max_delta_chain = 200
675
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
676
key_elements=keylength)
677
stream = transport.open_write_stream('newpack')
678
writer = pack.ContainerWriter(stream.write)
680
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
681
deltas=delta, add_callback=graph_index.add_nodes)
682
access = _DirectPackAccess({})
683
access.set_writer(writer, graph_index, (transport, 'newpack'))
684
result = KnitVersionedFiles(index, access,
685
max_delta_chain=max_delta_chain)
686
result.stream = stream
687
result.writer = writer
692
def cleanup_pack_knit(versioned_files):
693
versioned_files.stream.close()
694
versioned_files.writer.end()
697
class KnitVersionedFiles(VersionedFiles):
698
"""Storage for many versioned files using knit compression.
700
Backend storage is managed by indices and data objects.
702
:ivar _index: A _KnitGraphIndex or similar that can describe the
703
parents, graph, compression and data location of entries in this
704
KnitVersionedFiles. Note that this is only the index for
705
*this* vfs; if there are fallbacks they must be queried separately.
708
def __init__(self, index, data_access, max_delta_chain=200,
710
"""Create a KnitVersionedFiles with index and data_access.
712
:param index: The index for the knit data.
713
:param data_access: The access object to store and retrieve knit
715
:param max_delta_chain: The maximum number of deltas to permit during
716
insertion. Set to 0 to prohibit the use of deltas.
717
:param annotated: Set to True to cause annotations to be calculated and
718
stored during insertion.
721
self._access = data_access
722
self._max_delta_chain = max_delta_chain
724
self._factory = KnitAnnotateFactory()
726
self._factory = KnitPlainFactory()
727
self._fallback_vfs = []
729
def add_fallback_versioned_files(self, a_versioned_files):
730
"""Add a source of texts for texts not present in this knit.
732
:param a_versioned_files: A VersionedFiles object.
734
self._fallback_vfs.append(a_versioned_files)
736
def add_lines(self, key, parents, lines, parent_texts=None,
737
left_matching_blocks=None, nostore_sha=None, random_id=False,
739
"""See VersionedFiles.add_lines()."""
740
self._index._check_write_ok()
741
self._check_add(key, lines, random_id, check_content)
743
# The caller might pass None if there is no graph data, but kndx
744
# indexes can't directly store that, so we give them
745
# an empty tuple instead.
747
return self._add(key, lines, parents,
748
parent_texts, left_matching_blocks, nostore_sha, random_id)
750
def _add(self, key, lines, parents, parent_texts,
751
left_matching_blocks, nostore_sha, random_id):
752
"""Add a set of lines on top of version specified by parents.
754
Any versions not present will be converted into ghosts.
756
# first thing, if the content is something we don't need to store, find
758
line_bytes = ''.join(lines)
759
digest = sha_string(line_bytes)
760
if nostore_sha == digest:
761
raise errors.ExistingContent
764
if parent_texts is None:
766
# Do a single query to ascertain parent presence.
767
present_parent_map = self.get_parent_map(parents)
768
for parent in parents:
769
if parent in present_parent_map:
770
present_parents.append(parent)
772
# Currently we can only compress against the left most present parent.
773
if (len(present_parents) == 0 or
774
present_parents[0] != parents[0]):
777
# To speed the extract of texts the delta chain is limited
778
# to a fixed number of deltas. This should minimize both
779
# I/O and the time spend applying deltas.
780
delta = self._check_should_delta(present_parents[0])
782
text_length = len(line_bytes)
785
if lines[-1][-1] != '\n':
786
# copy the contents of lines.
788
options.append('no-eol')
789
lines[-1] = lines[-1] + '\n'
793
if type(element) != str:
794
raise TypeError("key contains non-strings: %r" % (key,))
795
# Knit hunks are still last-element only
797
content = self._factory.make(lines, version_id)
798
if 'no-eol' in options:
799
# Hint to the content object that its text() call should strip the
801
content._should_strip_eol = True
802
if delta or (self._factory.annotated and len(present_parents) > 0):
803
# Merge annotations from parent texts if needed.
804
delta_hunks = self._merge_annotations(content, present_parents,
805
parent_texts, delta, self._factory.annotated,
806
left_matching_blocks)
809
options.append('line-delta')
810
store_lines = self._factory.lower_line_delta(delta_hunks)
811
size, bytes = self._record_to_data(key, digest,
814
options.append('fulltext')
815
# isinstance is slower and we have no hierarchy.
816
if self._factory.__class__ == KnitPlainFactory:
817
# Use the already joined bytes saving iteration time in
819
size, bytes = self._record_to_data(key, digest,
822
# get mixed annotation + content and feed it into the
824
store_lines = self._factory.lower_fulltext(content)
825
size, bytes = self._record_to_data(key, digest,
828
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
829
self._index.add_records(
830
((key, options, access_memo, parents),),
832
return digest, text_length, content
834
def annotate(self, key):
835
"""See VersionedFiles.annotate."""
836
return self._factory.annotate(self, key)
838
def check(self, progress_bar=None):
839
"""See VersionedFiles.check()."""
840
# This doesn't actually test extraction of everything, but that will
841
# impact 'bzr check' substantially, and needs to be integrated with
842
# care. However, it does check for the obvious problem of a delta with
844
keys = self._index.keys()
845
parent_map = self.get_parent_map(keys)
847
if self._index.get_method(key) != 'fulltext':
848
compression_parent = parent_map[key][0]
849
if compression_parent not in parent_map:
850
raise errors.KnitCorrupt(self,
851
"Missing basis parent %s for %s" % (
852
compression_parent, key))
853
for fallback_vfs in self._fallback_vfs:
856
def _check_add(self, key, lines, random_id, check_content):
857
"""check that version_id and lines are safe to add."""
859
if contains_whitespace(version_id):
860
raise InvalidRevisionId(version_id, self)
861
self.check_not_reserved_id(version_id)
862
# TODO: If random_id==False and the key is already present, we should
863
# probably check that the existing content is identical to what is
864
# being inserted, and otherwise raise an exception. This would make
865
# the bundle code simpler.
867
self._check_lines_not_unicode(lines)
868
self._check_lines_are_lines(lines)
870
def _check_header(self, key, line):
871
rec = self._split_header(line)
872
self._check_header_version(rec, key[-1])
875
def _check_header_version(self, rec, version_id):
876
"""Checks the header version on original format knit records.
878
These have the last component of the key embedded in the record.
880
if rec[1] != version_id:
881
raise KnitCorrupt(self,
882
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
884
def _check_should_delta(self, parent):
885
"""Iterate back through the parent listing, looking for a fulltext.
887
This is used when we want to decide whether to add a delta or a new
888
fulltext. It searches for _max_delta_chain parents. When it finds a
889
fulltext parent, it sees if the total size of the deltas leading up to
890
it is large enough to indicate that we want a new full text anyway.
892
Return True if we should create a new delta, False if we should use a
897
for count in xrange(self._max_delta_chain):
898
# XXX: Collapse these two queries:
900
# Note that this only looks in the index of this particular
901
# KnitVersionedFiles, not in the fallbacks. This ensures that
902
# we won't store a delta spanning physical repository
904
method = self._index.get_method(parent)
905
except RevisionNotPresent:
906
# Some basis is not locally present: always delta
908
index, pos, size = self._index.get_position(parent)
909
if method == 'fulltext':
913
# We don't explicitly check for presence because this is in an
914
# inner loop, and if it's missing it'll fail anyhow.
915
# TODO: This should be asking for compression parent, not graph
917
parent = self._index.get_parent_map([parent])[parent][0]
919
# We couldn't find a fulltext, so we must create a new one
921
# Simple heuristic - if the total I/O wold be greater as a delta than
922
# the originally installed fulltext, we create a new fulltext.
923
return fulltext_size > delta_size
925
def _build_details_to_components(self, build_details):
926
"""Convert a build_details tuple to a position tuple."""
927
# record_details, access_memo, compression_parent
928
return build_details[3], build_details[0], build_details[1]
930
def _get_components_positions(self, keys, allow_missing=False):
931
"""Produce a map of position data for the components of keys.
933
This data is intended to be used for retrieving the knit records.
935
A dict of key to (record_details, index_memo, next, parents) is
937
method is the way referenced data should be applied.
938
index_memo is the handle to pass to the data access to actually get the
940
next is the build-parent of the version, or None for fulltexts.
941
parents is the version_ids of the parents of this version
943
:param allow_missing: If True do not raise an error on a missing component,
947
pending_components = keys
948
while pending_components:
949
build_details = self._index.get_build_details(pending_components)
950
current_components = set(pending_components)
951
pending_components = set()
952
for key, details in build_details.iteritems():
953
(index_memo, compression_parent, parents,
954
record_details) = details
955
method = record_details[0]
956
if compression_parent is not None:
957
pending_components.add(compression_parent)
958
component_data[key] = self._build_details_to_components(details)
959
missing = current_components.difference(build_details)
960
if missing and not allow_missing:
961
raise errors.RevisionNotPresent(missing.pop(), self)
962
return component_data
964
def _get_content(self, key, parent_texts={}):
965
"""Returns a content object that makes up the specified
967
cached_version = parent_texts.get(key, None)
968
if cached_version is not None:
969
# Ensure the cache dict is valid.
970
if not self.get_parent_map([key]):
971
raise RevisionNotPresent(key, self)
972
return cached_version
973
text_map, contents_map = self._get_content_maps([key])
974
return contents_map[key]
976
def _get_content_maps(self, keys, nonlocal_keys=None):
977
"""Produce maps of text and KnitContents
979
:param keys: The keys to produce content maps for.
980
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
981
which are known to not be in this knit, but rather in one of the
983
:return: (text_map, content_map) where text_map contains the texts for
984
the requested versions and content_map contains the KnitContents.
986
# FUTURE: This function could be improved for the 'extract many' case
987
# by tracking each component and only doing the copy when the number of
988
# children than need to apply delta's to it is > 1 or it is part of the
991
multiple_versions = len(keys) != 1
992
record_map = self._get_record_map(keys, allow_missing=True)
997
if nonlocal_keys is None:
998
nonlocal_keys = set()
1000
nonlocal_keys = frozenset(nonlocal_keys)
1001
missing_keys = set(nonlocal_keys)
1002
for source in self._fallback_vfs:
1003
if not missing_keys:
1005
for record in source.get_record_stream(missing_keys,
1007
if record.storage_kind == 'absent':
1009
missing_keys.remove(record.key)
1010
lines = split_lines(record.get_bytes_as('fulltext'))
1011
text_map[record.key] = lines
1012
content_map[record.key] = PlainKnitContent(lines, record.key)
1013
if record.key in keys:
1014
final_content[record.key] = content_map[record.key]
1016
if key in nonlocal_keys:
1021
while cursor is not None:
1023
record, record_details, digest, next = record_map[cursor]
1025
raise RevisionNotPresent(cursor, self)
1026
components.append((cursor, record, record_details, digest))
1028
if cursor in content_map:
1029
# no need to plan further back
1030
components.append((cursor, None, None, None))
1034
for (component_id, record, record_details,
1035
digest) in reversed(components):
1036
if component_id in content_map:
1037
content = content_map[component_id]
1039
content, delta = self._factory.parse_record(key[-1],
1040
record, record_details, content,
1041
copy_base_content=multiple_versions)
1042
if multiple_versions:
1043
content_map[component_id] = content
1045
final_content[key] = content
1047
# digest here is the digest from the last applied component.
1048
text = content.text()
1049
actual_sha = sha_strings(text)
1050
if actual_sha != digest:
1051
raise KnitCorrupt(self,
1053
'\n of reconstructed text does not match'
1055
'\n for version %s' %
1056
(actual_sha, digest, key))
1057
text_map[key] = text
1058
return text_map, final_content
1060
def get_parent_map(self, keys):
1061
"""Get a map of the graph parents of keys.
1063
:param keys: The keys to look up parents for.
1064
:return: A mapping from keys to parents. Absent keys are absent from
1067
return self._get_parent_map_with_sources(keys)[0]
1069
def _get_parent_map_with_sources(self, keys):
1070
"""Get a map of the parents of keys.
1072
:param keys: The keys to look up parents for.
1073
:return: A tuple. The first element is a mapping from keys to parents.
1074
Absent keys are absent from the mapping. The second element is a
1075
list with the locations each key was found in. The first element
1076
is the in-this-knit parents, the second the first fallback source,
1080
sources = [self._index] + self._fallback_vfs
1083
for source in sources:
1086
new_result = source.get_parent_map(missing)
1087
source_results.append(new_result)
1088
result.update(new_result)
1089
missing.difference_update(set(new_result))
1090
return result, source_results
1092
def _get_record_map(self, keys, allow_missing=False):
1093
"""Produce a dictionary of knit records.
1095
:return: {key:(record, record_details, digest, next)}
1097
data returned from read_records
1099
opaque information to pass to parse_record
1101
SHA1 digest of the full text after all steps are done
1103
build-parent of the version, i.e. the leftmost ancestor.
1104
Will be None if the record is not a delta.
1105
:param keys: The keys to build a map for
1106
:param allow_missing: If some records are missing, rather than
1107
error, just return the data that could be generated.
1109
position_map = self._get_components_positions(keys,
1110
allow_missing=allow_missing)
1111
# key = component_id, r = record_details, i_m = index_memo, n = next
1112
records = [(key, i_m) for key, (r, i_m, n)
1113
in position_map.iteritems()]
1115
for key, record, digest in \
1116
self._read_records_iter(records):
1117
(record_details, index_memo, next) = position_map[key]
1118
record_map[key] = record, record_details, digest, next
1121
def get_record_stream(self, keys, ordering, include_delta_closure):
1122
"""Get a stream of records for keys.
1124
:param keys: The keys to include.
1125
:param ordering: Either 'unordered' or 'topological'. A topologically
1126
sorted stream has compression parents strictly before their
1128
:param include_delta_closure: If True then the closure across any
1129
compression parents will be included (in the opaque data).
1130
:return: An iterator of ContentFactory objects, each of which is only
1131
valid until the iterator is advanced.
1133
# keys might be a generator
1137
if not self._index.has_graph:
1138
# Cannot topological order when no graph has been stored.
1139
ordering = 'unordered'
1140
if include_delta_closure:
1141
positions = self._get_components_positions(keys, allow_missing=True)
1143
build_details = self._index.get_build_details(keys)
1145
# (record_details, access_memo, compression_parent_key)
1146
positions = dict((key, self._build_details_to_components(details))
1147
for key, details in build_details.iteritems())
1148
absent_keys = keys.difference(set(positions))
1149
# There may be more absent keys : if we're missing the basis component
1150
# and are trying to include the delta closure.
1151
if include_delta_closure:
1152
needed_from_fallback = set()
1153
# Build up reconstructable_keys dict. key:True in this dict means
1154
# the key can be reconstructed.
1155
reconstructable_keys = {}
1159
chain = [key, positions[key][2]]
1161
needed_from_fallback.add(key)
1164
while chain[-1] is not None:
1165
if chain[-1] in reconstructable_keys:
1166
result = reconstructable_keys[chain[-1]]
1170
chain.append(positions[chain[-1]][2])
1172
# missing basis component
1173
needed_from_fallback.add(chain[-1])
1176
for chain_key in chain[:-1]:
1177
reconstructable_keys[chain_key] = result
1179
needed_from_fallback.add(key)
1180
# Double index lookups here : need a unified api ?
1181
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1182
if ordering == 'topological':
1183
# Global topological sort
1184
present_keys = tsort.topo_sort(global_map)
1185
# Now group by source:
1187
current_source = None
1188
for key in present_keys:
1189
for parent_map in parent_maps:
1190
if key in parent_map:
1191
key_source = parent_map
1193
if current_source is not key_source:
1194
source_keys.append((key_source, []))
1195
current_source = key_source
1196
source_keys[-1][1].append(key)
1198
if ordering != 'unordered':
1199
raise AssertionError('valid values for ordering are:'
1200
' "unordered" or "topological" not: %r'
1202
# Just group by source; remote sources first.
1205
for parent_map in reversed(parent_maps):
1206
source_keys.append((parent_map, []))
1207
for key in parent_map:
1208
present_keys.append(key)
1209
source_keys[-1][1].append(key)
1210
absent_keys = keys - set(global_map)
1211
for key in absent_keys:
1212
yield AbsentContentFactory(key)
1213
# restrict our view to the keys we can answer.
1214
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1215
# XXX: At that point we need to consider the impact of double reads by
1216
# utilising components multiple times.
1217
if include_delta_closure:
1218
# XXX: get_content_maps performs its own index queries; allow state
1220
text_map, _ = self._get_content_maps(present_keys,
1221
needed_from_fallback - absent_keys)
1222
for key in present_keys:
1223
yield FulltextContentFactory(key, global_map[key], None,
1224
''.join(text_map[key]))
1226
for source, keys in source_keys:
1227
if source is parent_maps[0]:
1228
# this KnitVersionedFiles
1229
records = [(key, positions[key][1]) for key in keys]
1230
for key, raw_data, sha1 in self._read_records_iter_raw(records):
1231
(record_details, index_memo, _) = positions[key]
1232
yield KnitContentFactory(key, global_map[key],
1233
record_details, sha1, raw_data, self._factory.annotated, None)
1235
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1236
for record in vf.get_record_stream(keys, ordering,
1237
include_delta_closure):
1240
def get_sha1s(self, keys):
1241
"""See VersionedFiles.get_sha1s()."""
1243
record_map = self._get_record_map(missing, allow_missing=True)
1245
for key, details in record_map.iteritems():
1246
if key not in missing:
1248
# record entry 2 is the 'digest'.
1249
result[key] = details[2]
1250
missing.difference_update(set(result))
1251
for source in self._fallback_vfs:
1254
new_result = source.get_sha1s(missing)
1255
result.update(new_result)
1256
missing.difference_update(set(new_result))
1259
def insert_record_stream(self, stream):
1260
"""Insert a record stream into this container.
1262
:param stream: A stream of records to insert.
1264
:seealso VersionedFiles.get_record_stream:
1266
def get_adapter(adapter_key):
1268
return adapters[adapter_key]
1270
adapter_factory = adapter_registry.get(adapter_key)
1271
adapter = adapter_factory(self)
1272
adapters[adapter_key] = adapter
1274
if self._factory.annotated:
1275
# self is annotated, we need annotated knits to use directly.
1276
annotated = "annotated-"
1279
# self is not annotated, but we can strip annotations cheaply.
1281
convertibles = set(["knit-annotated-ft-gz"])
1282
if self._max_delta_chain:
1283
convertibles.add("knit-annotated-delta-gz")
1284
# The set of types we can cheaply adapt without needing basis texts.
1285
native_types = set()
1286
if self._max_delta_chain:
1287
native_types.add("knit-%sdelta-gz" % annotated)
1288
native_types.add("knit-%sft-gz" % annotated)
1289
knit_types = native_types.union(convertibles)
1291
# Buffer all index entries that we can't add immediately because their
1292
# basis parent is missing. We don't buffer all because generating
1293
# annotations may require access to some of the new records. However we
1294
# can't generate annotations from new deltas until their basis parent
1295
# is present anyway, so we get away with not needing an index that
1296
# includes the new keys.
1297
# key = basis_parent, value = index entry to add
1298
buffered_index_entries = {}
1299
for record in stream:
1300
parents = record.parents
1301
# Raise an error when a record is missing.
1302
if record.storage_kind == 'absent':
1303
raise RevisionNotPresent([record.key], self)
1304
if record.storage_kind in knit_types:
1305
if record.storage_kind not in native_types:
1307
adapter_key = (record.storage_kind, "knit-delta-gz")
1308
adapter = get_adapter(adapter_key)
1310
adapter_key = (record.storage_kind, "knit-ft-gz")
1311
adapter = get_adapter(adapter_key)
1312
bytes = adapter.get_bytes(
1313
record, record.get_bytes_as(record.storage_kind))
1315
bytes = record.get_bytes_as(record.storage_kind)
1316
options = [record._build_details[0]]
1317
if record._build_details[1]:
1318
options.append('no-eol')
1319
# Just blat it across.
1320
# Note: This does end up adding data on duplicate keys. As
1321
# modern repositories use atomic insertions this should not
1322
# lead to excessive growth in the event of interrupted fetches.
1323
# 'knit' repositories may suffer excessive growth, but as a
1324
# deprecated format this is tolerable. It can be fixed if
1325
# needed by in the kndx index support raising on a duplicate
1326
# add with identical parents and options.
1327
access_memo = self._access.add_raw_records(
1328
[(record.key, len(bytes))], bytes)[0]
1329
index_entry = (record.key, options, access_memo, parents)
1331
if 'fulltext' not in options:
1332
basis_parent = parents[0]
1333
# Note that pack backed knits don't need to buffer here
1334
# because they buffer all writes to the transaction level,
1335
# but we don't expose that difference at the index level. If
1336
# the query here has sufficient cost to show up in
1337
# profiling we should do that.
1338
if basis_parent not in self.get_parent_map([basis_parent]):
1339
pending = buffered_index_entries.setdefault(
1341
pending.append(index_entry)
1344
self._index.add_records([index_entry])
1345
elif record.storage_kind == 'fulltext':
1346
self.add_lines(record.key, parents,
1347
split_lines(record.get_bytes_as('fulltext')))
1349
adapter_key = record.storage_kind, 'fulltext'
1350
adapter = get_adapter(adapter_key)
1351
lines = split_lines(adapter.get_bytes(
1352
record, record.get_bytes_as(record.storage_kind)))
1354
self.add_lines(record.key, parents, lines)
1355
except errors.RevisionAlreadyPresent:
1357
# Add any records whose basis parent is now available.
1358
added_keys = [record.key]
1360
key = added_keys.pop(0)
1361
if key in buffered_index_entries:
1362
index_entries = buffered_index_entries[key]
1363
self._index.add_records(index_entries)
1365
[index_entry[0] for index_entry in index_entries])
1366
del buffered_index_entries[key]
1367
# If there were any deltas which had a missing basis parent, error.
1368
if buffered_index_entries:
1369
raise errors.RevisionNotPresent(buffered_index_entries.keys()[0],
1372
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1373
"""Iterate over the lines in the versioned files from keys.
1375
This may return lines from other keys. Each item the returned
1376
iterator yields is a tuple of a line and a text version that that line
1377
is present in (not introduced in).
1379
Ordering of results is in whatever order is most suitable for the
1380
underlying storage format.
1382
If a progress bar is supplied, it may be used to indicate progress.
1383
The caller is responsible for cleaning up progress bars (because this
1387
* Lines are normalised by the underlying store: they will all have \n
1389
* Lines are returned in arbitrary order.
1391
:return: An iterator over (line, key).
1394
pb = progress.DummyProgress()
1397
# we don't care about inclusions, the caller cares.
1398
# but we need to setup a list of records to visit.
1399
# we need key, position, length
1401
build_details = self._index.get_build_details(keys)
1402
for key, details in build_details.iteritems():
1404
key_records.append((key, details[0]))
1406
records_iter = enumerate(self._read_records_iter(key_records))
1407
for (key_idx, (key, data, sha_value)) in records_iter:
1408
pb.update('Walking content.', key_idx, total)
1409
compression_parent = build_details[key][1]
1410
if compression_parent is None:
1412
line_iterator = self._factory.get_fulltext_content(data)
1415
line_iterator = self._factory.get_linedelta_content(data)
1416
# XXX: It might be more efficient to yield (key,
1417
# line_iterator) in the future. However for now, this is a simpler
1418
# change to integrate into the rest of the codebase. RBC 20071110
1419
for line in line_iterator:
1421
for source in self._fallback_vfs:
1425
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1426
source_keys.add(key)
1428
keys.difference_update(source_keys)
1430
raise RevisionNotPresent(keys, self.filename)
1431
pb.update('Walking content.', total, total)
1433
def _make_line_delta(self, delta_seq, new_content):
1434
"""Generate a line delta from delta_seq and new_content."""
1436
for op in delta_seq.get_opcodes():
1437
if op[0] == 'equal':
1439
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1442
def _merge_annotations(self, content, parents, parent_texts={},
1443
delta=None, annotated=None,
1444
left_matching_blocks=None):
1445
"""Merge annotations for content and generate deltas.
1447
This is done by comparing the annotations based on changes to the text
1448
and generating a delta on the resulting full texts. If annotations are
1449
not being created then a simple delta is created.
1451
if left_matching_blocks is not None:
1452
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1456
for parent_key in parents:
1457
merge_content = self._get_content(parent_key, parent_texts)
1458
if (parent_key == parents[0] and delta_seq is not None):
1461
seq = patiencediff.PatienceSequenceMatcher(
1462
None, merge_content.text(), content.text())
1463
for i, j, n in seq.get_matching_blocks():
1466
# this copies (origin, text) pairs across to the new
1467
# content for any line that matches the last-checked
1469
content._lines[j:j+n] = merge_content._lines[i:i+n]
1470
# XXX: Robert says the following block is a workaround for a
1471
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1472
if content._lines and content._lines[-1][1][-1] != '\n':
1473
# The copied annotation was from a line without a trailing EOL,
1474
# reinstate one for the content object, to ensure correct
1476
line = content._lines[-1][1] + '\n'
1477
content._lines[-1] = (content._lines[-1][0], line)
1479
if delta_seq is None:
1480
reference_content = self._get_content(parents[0], parent_texts)
1481
new_texts = content.text()
1482
old_texts = reference_content.text()
1483
delta_seq = patiencediff.PatienceSequenceMatcher(
1484
None, old_texts, new_texts)
1485
return self._make_line_delta(delta_seq, content)
1487
def _parse_record(self, version_id, data):
1488
"""Parse an original format knit record.
1490
These have the last element of the key only present in the stored data.
1492
rec, record_contents = self._parse_record_unchecked(data)
1493
self._check_header_version(rec, version_id)
1494
return record_contents, rec[3]
1496
def _parse_record_header(self, key, raw_data):
1497
"""Parse a record header for consistency.
1499
:return: the header and the decompressor stream.
1500
as (stream, header_record)
1502
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1505
rec = self._check_header(key, df.readline())
1506
except Exception, e:
1507
raise KnitCorrupt(self,
1508
"While reading {%s} got %s(%s)"
1509
% (key, e.__class__.__name__, str(e)))
1512
def _parse_record_unchecked(self, data):
1514
# 4168 calls in 2880 217 internal
1515
# 4168 calls to _parse_record_header in 2121
1516
# 4168 calls to readlines in 330
1517
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1519
record_contents = df.readlines()
1520
except Exception, e:
1521
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1522
(data, e.__class__.__name__, str(e)))
1523
header = record_contents.pop(0)
1524
rec = self._split_header(header)
1525
last_line = record_contents.pop()
1526
if len(record_contents) != int(rec[2]):
1527
raise KnitCorrupt(self,
1528
'incorrect number of lines %s != %s'
1529
' for version {%s} %s'
1530
% (len(record_contents), int(rec[2]),
1531
rec[1], record_contents))
1532
if last_line != 'end %s\n' % rec[1]:
1533
raise KnitCorrupt(self,
1534
'unexpected version end line %r, wanted %r'
1535
% (last_line, rec[1]))
1537
return rec, record_contents
1539
def _read_records_iter(self, records):
1540
"""Read text records from data file and yield result.
1542
The result will be returned in whatever is the fastest to read.
1543
Not by the order requested. Also, multiple requests for the same
1544
record will only yield 1 response.
1545
:param records: A list of (key, access_memo) entries
1546
:return: Yields (key, contents, digest) in the order
1547
read, not the order requested
1552
# XXX: This smells wrong, IO may not be getting ordered right.
1553
needed_records = sorted(set(records), key=operator.itemgetter(1))
1554
if not needed_records:
1557
# The transport optimizes the fetching as well
1558
# (ie, reads continuous ranges.)
1559
raw_data = self._access.get_raw_records(
1560
[index_memo for key, index_memo in needed_records])
1562
for (key, index_memo), data in \
1563
izip(iter(needed_records), raw_data):
1564
content, digest = self._parse_record(key[-1], data)
1565
yield key, content, digest
1567
def _read_records_iter_raw(self, records):
1568
"""Read text records from data file and yield raw data.
1570
This unpacks enough of the text record to validate the id is
1571
as expected but thats all.
1573
Each item the iterator yields is (key, bytes, sha1_of_full_text).
1575
# setup an iterator of the external records:
1576
# uses readv so nice and fast we hope.
1578
# grab the disk data needed.
1579
needed_offsets = [index_memo for key, index_memo
1581
raw_records = self._access.get_raw_records(needed_offsets)
1583
for key, index_memo in records:
1584
data = raw_records.next()
1585
# validate the header (note that we can only use the suffix in
1586
# current knit records).
1587
df, rec = self._parse_record_header(key, data)
1589
yield key, data, rec[3]
1591
def _record_to_data(self, key, digest, lines, dense_lines=None):
1592
"""Convert key, digest, lines into a raw data block.
1594
:param key: The key of the record. Currently keys are always serialised
1595
using just the trailing component.
1596
:param dense_lines: The bytes of lines but in a denser form. For
1597
instance, if lines is a list of 1000 bytestrings each ending in \n,
1598
dense_lines may be a list with one line in it, containing all the
1599
1000's lines and their \n's. Using dense_lines if it is already
1600
known is a win because the string join to create bytes in this
1601
function spends less time resizing the final string.
1602
:return: (len, a StringIO instance with the raw data ready to read.)
1604
# Note: using a string copy here increases memory pressure with e.g.
1605
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
1606
# when doing the initial commit of a mozilla tree. RBC 20070921
1607
bytes = ''.join(chain(
1608
["version %s %d %s\n" % (key[-1],
1611
dense_lines or lines,
1612
["end %s\n" % key[-1]]))
1613
if type(bytes) != str:
1614
raise AssertionError(
1615
'data must be plain bytes was %s' % type(bytes))
1616
if lines and lines[-1][-1] != '\n':
1617
raise ValueError('corrupt lines value %r' % lines)
1618
compressed_bytes = tuned_gzip.bytes_to_gzip(bytes)
1619
return len(compressed_bytes), compressed_bytes
1621
def _split_header(self, line):
1624
raise KnitCorrupt(self,
1625
'unexpected number of elements in record header')
1629
"""See VersionedFiles.keys."""
1630
if 'evil' in debug.debug_flags:
1631
trace.mutter_callsite(2, "keys scales with size of history")
1632
sources = [self._index] + self._fallback_vfs
1634
for source in sources:
1635
result.update(source.keys())
1640
class _KndxIndex(object):
1641
"""Manages knit index files
1643
The index is kept in memory and read on startup, to enable
1644
fast lookups of revision information. The cursor of the index
1645
file is always pointing to the end, making it easy to append
1648
_cache is a cache for fast mapping from version id to a Index
1651
_history is a cache for fast mapping from indexes to version ids.
1653
The index data format is dictionary compressed when it comes to
1654
parent references; a index entry may only have parents that with a
1655
lover index number. As a result, the index is topological sorted.
1657
Duplicate entries may be written to the index for a single version id
1658
if this is done then the latter one completely replaces the former:
1659
this allows updates to correct version and parent information.
1660
Note that the two entries may share the delta, and that successive
1661
annotations and references MUST point to the first entry.
1663
The index file on disc contains a header, followed by one line per knit
1664
record. The same revision can be present in an index file more than once.
1665
The first occurrence gets assigned a sequence number starting from 0.
1667
The format of a single line is
1668
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1669
REVISION_ID is a utf8-encoded revision id
1670
FLAGS is a comma separated list of flags about the record. Values include
1671
no-eol, line-delta, fulltext.
1672
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1673
that the the compressed data starts at.
1674
LENGTH is the ascii representation of the length of the data file.
1675
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1677
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1678
revision id already in the knit that is a parent of REVISION_ID.
1679
The ' :' marker is the end of record marker.
1682
when a write is interrupted to the index file, it will result in a line
1683
that does not end in ' :'. If the ' :' is not present at the end of a line,
1684
or at the end of the file, then the record that is missing it will be
1685
ignored by the parser.
1687
When writing new records to the index file, the data is preceded by '\n'
1688
to ensure that records always start on new lines even if the last write was
1689
interrupted. As a result its normal for the last line in the index to be
1690
missing a trailing newline. One can be added with no harmful effects.
1692
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
1693
where prefix is e.g. the (fileid,) for .texts instances or () for
1694
constant-mapped things like .revisions, and the old state is
1695
tuple(history_vector, cache_dict). This is used to prevent having an
1696
ABI change with the C extension that reads .kndx files.
1699
HEADER = "# bzr knit index 8\n"
1701
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
1702
"""Create a _KndxIndex on transport using mapper."""
1703
self._transport = transport
1704
self._mapper = mapper
1705
self._get_scope = get_scope
1706
self._allow_writes = allow_writes
1707
self._is_locked = is_locked
1709
self.has_graph = True
1711
def add_records(self, records, random_id=False):
1712
"""Add multiple records to the index.
1714
:param records: a list of tuples:
1715
(key, options, access_memo, parents).
1716
:param random_id: If True the ids being added were randomly generated
1717
and no check for existence will be performed.
1720
for record in records:
1723
path = self._mapper.map(key) + '.kndx'
1724
path_keys = paths.setdefault(path, (prefix, []))
1725
path_keys[1].append(record)
1726
for path in sorted(paths):
1727
prefix, path_keys = paths[path]
1728
self._load_prefixes([prefix])
1730
orig_history = self._kndx_cache[prefix][1][:]
1731
orig_cache = self._kndx_cache[prefix][0].copy()
1734
for key, options, (_, pos, size), parents in path_keys:
1736
# kndx indices cannot be parentless.
1738
line = "\n%s %s %s %s %s :" % (
1739
key[-1], ','.join(options), pos, size,
1740
self._dictionary_compress(parents))
1741
if type(line) != str:
1742
raise AssertionError(
1743
'data must be utf8 was %s' % type(line))
1745
self._cache_key(key, options, pos, size, parents)
1746
if len(orig_history):
1747
self._transport.append_bytes(path, ''.join(lines))
1749
self._init_index(path, lines)
1751
# If any problems happen, restore the original values and re-raise
1752
self._kndx_cache[prefix] = (orig_cache, orig_history)
1755
def _cache_key(self, key, options, pos, size, parent_keys):
1756
"""Cache a version record in the history array and index cache.
1758
This is inlined into _load_data for performance. KEEP IN SYNC.
1759
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1763
version_id = key[-1]
1764
# last-element only for compatibilty with the C load_data.
1765
parents = tuple(parent[-1] for parent in parent_keys)
1766
for parent in parent_keys:
1767
if parent[:-1] != prefix:
1768
raise ValueError("mismatched prefixes for %r, %r" % (
1770
cache, history = self._kndx_cache[prefix]
1771
# only want the _history index to reference the 1st index entry
1773
if version_id not in cache:
1774
index = len(history)
1775
history.append(version_id)
1777
index = cache[version_id][5]
1778
cache[version_id] = (version_id,
1785
def check_header(self, fp):
1786
line = fp.readline()
1788
# An empty file can actually be treated as though the file doesn't
1790
raise errors.NoSuchFile(self)
1791
if line != self.HEADER:
1792
raise KnitHeaderError(badline=line, filename=self)
1794
def _check_read(self):
1795
if not self._is_locked():
1796
raise errors.ObjectNotLocked(self)
1797
if self._get_scope() != self._scope:
1800
def _check_write_ok(self):
1801
"""Assert if not writes are permitted."""
1802
if not self._is_locked():
1803
raise errors.ObjectNotLocked(self)
1804
if self._get_scope() != self._scope:
1806
if self._mode != 'w':
1807
raise errors.ReadOnlyObjectDirtiedError(self)
1809
def get_build_details(self, keys):
1810
"""Get the method, index_memo and compression parent for keys.
1812
Ghosts are omitted from the result.
1814
:param keys: An iterable of keys.
1815
:return: A dict of key:(index_memo, compression_parent, parents,
1818
opaque structure to pass to read_records to extract the raw
1821
Content that this record is built upon, may be None
1823
Logical parents of this node
1825
extra information about the content which needs to be passed to
1826
Factory.parse_record
1828
prefixes = self._partition_keys(keys)
1829
parent_map = self.get_parent_map(keys)
1832
if key not in parent_map:
1834
method = self.get_method(key)
1835
parents = parent_map[key]
1836
if method == 'fulltext':
1837
compression_parent = None
1839
compression_parent = parents[0]
1840
noeol = 'no-eol' in self.get_options(key)
1841
index_memo = self.get_position(key)
1842
result[key] = (index_memo, compression_parent,
1843
parents, (method, noeol))
1846
def get_method(self, key):
1847
"""Return compression method of specified key."""
1848
options = self.get_options(key)
1849
if 'fulltext' in options:
1851
elif 'line-delta' in options:
1854
raise errors.KnitIndexUnknownMethod(self, options)
1856
def get_options(self, key):
1857
"""Return a list representing options.
1861
prefix, suffix = self._split_key(key)
1862
self._load_prefixes([prefix])
1864
return self._kndx_cache[prefix][0][suffix][1]
1866
raise RevisionNotPresent(key, self)
1868
def get_parent_map(self, keys):
1869
"""Get a map of the parents of keys.
1871
:param keys: The keys to look up parents for.
1872
:return: A mapping from keys to parents. Absent keys are absent from
1875
# Parse what we need to up front, this potentially trades off I/O
1876
# locality (.kndx and .knit in the same block group for the same file
1877
# id) for less checking in inner loops.
1878
prefixes = set(key[:-1] for key in keys)
1879
self._load_prefixes(prefixes)
1884
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
1888
result[key] = tuple(prefix + (suffix,) for
1889
suffix in suffix_parents)
1892
def get_position(self, key):
1893
"""Return details needed to access the version.
1895
:return: a tuple (key, data position, size) to hand to the access
1896
logic to get the record.
1898
prefix, suffix = self._split_key(key)
1899
self._load_prefixes([prefix])
1900
entry = self._kndx_cache[prefix][0][suffix]
1901
return key, entry[2], entry[3]
1903
def _init_index(self, path, extra_lines=[]):
1904
"""Initialize an index."""
1906
sio.write(self.HEADER)
1907
sio.writelines(extra_lines)
1909
self._transport.put_file_non_atomic(path, sio,
1910
create_parent_dir=True)
1911
# self._create_parent_dir)
1912
# mode=self._file_mode,
1913
# dir_mode=self._dir_mode)
1916
"""Get all the keys in the collection.
1918
The keys are not ordered.
1921
# Identify all key prefixes.
1922
# XXX: A bit hacky, needs polish.
1923
if type(self._mapper) == ConstantMapper:
1927
for quoted_relpath in self._transport.iter_files_recursive():
1928
path, ext = os.path.splitext(quoted_relpath)
1930
prefixes = [self._mapper.unmap(path) for path in relpaths]
1931
self._load_prefixes(prefixes)
1932
for prefix in prefixes:
1933
for suffix in self._kndx_cache[prefix][1]:
1934
result.add(prefix + (suffix,))
1937
def _load_prefixes(self, prefixes):
1938
"""Load the indices for prefixes."""
1940
for prefix in prefixes:
1941
if prefix not in self._kndx_cache:
1942
# the load_data interface writes to these variables.
1945
self._filename = prefix
1947
path = self._mapper.map(prefix) + '.kndx'
1948
fp = self._transport.get(path)
1950
# _load_data may raise NoSuchFile if the target knit is
1952
_load_data(self, fp)
1955
self._kndx_cache[prefix] = (self._cache, self._history)
1960
self._kndx_cache[prefix] = ({}, [])
1961
if type(self._mapper) == ConstantMapper:
1962
# preserve behaviour for revisions.kndx etc.
1963
self._init_index(path)
1968
def _partition_keys(self, keys):
1969
"""Turn keys into a dict of prefix:suffix_list."""
1972
prefix_keys = result.setdefault(key[:-1], [])
1973
prefix_keys.append(key[-1])
1976
def _dictionary_compress(self, keys):
1977
"""Dictionary compress keys.
1979
:param keys: The keys to generate references to.
1980
:return: A string representation of keys. keys which are present are
1981
dictionary compressed, and others are emitted as fulltext with a
1987
prefix = keys[0][:-1]
1988
cache = self._kndx_cache[prefix][0]
1990
if key[:-1] != prefix:
1991
# kndx indices cannot refer across partitioned storage.
1992
raise ValueError("mismatched prefixes for %r" % keys)
1993
if key[-1] in cache:
1994
# -- inlined lookup() --
1995
result_list.append(str(cache[key[-1]][5]))
1996
# -- end lookup () --
1998
result_list.append('.' + key[-1])
1999
return ' '.join(result_list)
2001
def _reset_cache(self):
2002
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2003
# (cache_dict, history_vector) for parsed kndx files.
2004
self._kndx_cache = {}
2005
self._scope = self._get_scope()
2006
allow_writes = self._allow_writes()
2012
def _split_key(self, key):
2013
"""Split key into a prefix and suffix."""
2014
return key[:-1], key[-1]
2017
class _KnitGraphIndex(object):
2018
"""A KnitVersionedFiles index layered on GraphIndex."""
2020
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2022
"""Construct a KnitGraphIndex on a graph_index.
2024
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2025
:param is_locked: A callback to check whether the object should answer
2027
:param deltas: Allow delta-compressed records.
2028
:param parents: If True, record knits parents, if not do not record
2030
:param add_callback: If not None, allow additions to the index and call
2031
this callback with a list of added GraphIndex nodes:
2032
[(node, value, node_refs), ...]
2033
:param is_locked: A callback, returns True if the index is locked and
2036
self._add_callback = add_callback
2037
self._graph_index = graph_index
2038
self._deltas = deltas
2039
self._parents = parents
2040
if deltas and not parents:
2041
# XXX: TODO: Delta tree and parent graph should be conceptually
2043
raise KnitCorrupt(self, "Cannot do delta compression without "
2045
self.has_graph = parents
2046
self._is_locked = is_locked
2049
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2051
def add_records(self, records, random_id=False):
2052
"""Add multiple records to the index.
2054
This function does not insert data into the Immutable GraphIndex
2055
backing the KnitGraphIndex, instead it prepares data for insertion by
2056
the caller and checks that it is safe to insert then calls
2057
self._add_callback with the prepared GraphIndex nodes.
2059
:param records: a list of tuples:
2060
(key, options, access_memo, parents).
2061
:param random_id: If True the ids being added were randomly generated
2062
and no check for existence will be performed.
2064
if not self._add_callback:
2065
raise errors.ReadOnlyError(self)
2066
# we hope there are no repositories with inconsistent parentage
2070
for (key, options, access_memo, parents) in records:
2072
parents = tuple(parents)
2073
index, pos, size = access_memo
2074
if 'no-eol' in options:
2078
value += "%d %d" % (pos, size)
2079
if not self._deltas:
2080
if 'line-delta' in options:
2081
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2084
if 'line-delta' in options:
2085
node_refs = (parents, (parents[0],))
2087
node_refs = (parents, ())
2089
node_refs = (parents, )
2092
raise KnitCorrupt(self, "attempt to add node with parents "
2093
"in parentless index.")
2095
keys[key] = (value, node_refs)
2098
present_nodes = self._get_entries(keys)
2099
for (index, key, value, node_refs) in present_nodes:
2100
if (value[0] != keys[key][0][0] or
2101
node_refs != keys[key][1]):
2102
raise KnitCorrupt(self, "inconsistent details in add_records"
2103
": %s %s" % ((value, node_refs), keys[key]))
2107
for key, (value, node_refs) in keys.iteritems():
2108
result.append((key, value, node_refs))
2110
for key, (value, node_refs) in keys.iteritems():
2111
result.append((key, value))
2112
self._add_callback(result)
2114
def _check_read(self):
2115
"""raise if reads are not permitted."""
2116
if not self._is_locked():
2117
raise errors.ObjectNotLocked(self)
2119
def _check_write_ok(self):
2120
"""Assert if writes are not permitted."""
2121
if not self._is_locked():
2122
raise errors.ObjectNotLocked(self)
2124
def _compression_parent(self, an_entry):
2125
# return the key that an_entry is compressed against, or None
2126
# Grab the second parent list (as deltas implies parents currently)
2127
compression_parents = an_entry[3][1]
2128
if not compression_parents:
2130
if len(compression_parents) != 1:
2131
raise AssertionError(
2132
"Too many compression parents: %r" % compression_parents)
2133
return compression_parents[0]
2135
def get_build_details(self, keys):
2136
"""Get the method, index_memo and compression parent for version_ids.
2138
Ghosts are omitted from the result.
2140
:param keys: An iterable of keys.
2141
:return: A dict of key:
2142
(index_memo, compression_parent, parents, record_details).
2144
opaque structure to pass to read_records to extract the raw
2147
Content that this record is built upon, may be None
2149
Logical parents of this node
2151
extra information about the content which needs to be passed to
2152
Factory.parse_record
2156
entries = self._get_entries(keys, False)
2157
for entry in entries:
2159
if not self._parents:
2162
parents = entry[3][0]
2163
if not self._deltas:
2164
compression_parent_key = None
2166
compression_parent_key = self._compression_parent(entry)
2167
noeol = (entry[2][0] == 'N')
2168
if compression_parent_key:
2169
method = 'line-delta'
2172
result[key] = (self._node_to_position(entry),
2173
compression_parent_key, parents,
2177
def _get_entries(self, keys, check_present=False):
2178
"""Get the entries for keys.
2180
:param keys: An iterable of index key tuples.
2185
for node in self._graph_index.iter_entries(keys):
2187
found_keys.add(node[1])
2189
# adapt parentless index to the rest of the code.
2190
for node in self._graph_index.iter_entries(keys):
2191
yield node[0], node[1], node[2], ()
2192
found_keys.add(node[1])
2194
missing_keys = keys.difference(found_keys)
2196
raise RevisionNotPresent(missing_keys.pop(), self)
2198
def get_method(self, key):
2199
"""Return compression method of specified key."""
2200
return self._get_method(self._get_node(key))
2202
def _get_method(self, node):
2203
if not self._deltas:
2205
if self._compression_parent(node):
2210
def _get_node(self, key):
2212
return list(self._get_entries([key]))[0]
2214
raise RevisionNotPresent(key, self)
2216
def get_options(self, key):
2217
"""Return a list representing options.
2221
node = self._get_node(key)
2222
options = [self._get_method(node)]
2223
if node[2][0] == 'N':
2224
options.append('no-eol')
2227
def get_parent_map(self, keys):
2228
"""Get a map of the parents of keys.
2230
:param keys: The keys to look up parents for.
2231
:return: A mapping from keys to parents. Absent keys are absent from
2235
nodes = self._get_entries(keys)
2239
result[node[1]] = node[3][0]
2242
result[node[1]] = None
2245
def get_position(self, key):
2246
"""Return details needed to access the version.
2248
:return: a tuple (index, data position, size) to hand to the access
2249
logic to get the record.
2251
node = self._get_node(key)
2252
return self._node_to_position(node)
2255
"""Get all the keys in the collection.
2257
The keys are not ordered.
2260
return [node[1] for node in self._graph_index.iter_all_entries()]
2262
def _node_to_position(self, node):
2263
"""Convert an index value to position details."""
2264
bits = node[2][1:].split(' ')
2265
return node[0], int(bits[0]), int(bits[1])
2268
class _KnitKeyAccess(object):
2269
"""Access to records in .knit files."""
2271
def __init__(self, transport, mapper):
2272
"""Create a _KnitKeyAccess with transport and mapper.
2274
:param transport: The transport the access object is rooted at.
2275
:param mapper: The mapper used to map keys to .knit files.
2277
self._transport = transport
2278
self._mapper = mapper
2280
def add_raw_records(self, key_sizes, raw_data):
2281
"""Add raw knit bytes to a storage area.
2283
The data is spooled to the container writer in one bytes-record per
2286
:param sizes: An iterable of tuples containing the key and size of each
2288
:param raw_data: A bytestring containing the data.
2289
:return: A list of memos to retrieve the record later. Each memo is an
2290
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
2291
length), where the key is the record key.
2293
if type(raw_data) != str:
2294
raise AssertionError(
2295
'data must be plain bytes was %s' % type(raw_data))
2298
# TODO: This can be tuned for writing to sftp and other servers where
2299
# append() is relatively expensive by grouping the writes to each key
2301
for key, size in key_sizes:
2302
path = self._mapper.map(key)
2304
base = self._transport.append_bytes(path + '.knit',
2305
raw_data[offset:offset+size])
2306
except errors.NoSuchFile:
2307
self._transport.mkdir(osutils.dirname(path))
2308
base = self._transport.append_bytes(path + '.knit',
2309
raw_data[offset:offset+size])
2313
result.append((key, base, size))
2316
def get_raw_records(self, memos_for_retrieval):
2317
"""Get the raw bytes for a records.
2319
:param memos_for_retrieval: An iterable containing the access memo for
2320
retrieving the bytes.
2321
:return: An iterator over the bytes of the records.
2323
# first pass, group into same-index request to minimise readv's issued.
2325
current_prefix = None
2326
for (key, offset, length) in memos_for_retrieval:
2327
if current_prefix == key[:-1]:
2328
current_list.append((offset, length))
2330
if current_prefix is not None:
2331
request_lists.append((current_prefix, current_list))
2332
current_prefix = key[:-1]
2333
current_list = [(offset, length)]
2334
# handle the last entry
2335
if current_prefix is not None:
2336
request_lists.append((current_prefix, current_list))
2337
for prefix, read_vector in request_lists:
2338
path = self._mapper.map(prefix) + '.knit'
2339
for pos, data in self._transport.readv(path, read_vector):
2343
class _DirectPackAccess(object):
2344
"""Access to data in one or more packs with less translation."""
2346
def __init__(self, index_to_packs):
2347
"""Create a _DirectPackAccess object.
2349
:param index_to_packs: A dict mapping index objects to the transport
2350
and file names for obtaining data.
2352
self._container_writer = None
2353
self._write_index = None
2354
self._indices = index_to_packs
2356
def add_raw_records(self, key_sizes, raw_data):
2357
"""Add raw knit bytes to a storage area.
2359
The data is spooled to the container writer in one bytes-record per
2362
:param sizes: An iterable of tuples containing the key and size of each
2364
:param raw_data: A bytestring containing the data.
2365
:return: A list of memos to retrieve the record later. Each memo is an
2366
opaque index memo. For _DirectPackAccess the memo is (index, pos,
2367
length), where the index field is the write_index object supplied
2368
to the PackAccess object.
2370
if type(raw_data) != str:
2371
raise AssertionError(
2372
'data must be plain bytes was %s' % type(raw_data))
2375
for key, size in key_sizes:
2376
p_offset, p_length = self._container_writer.add_bytes_record(
2377
raw_data[offset:offset+size], [])
2379
result.append((self._write_index, p_offset, p_length))
2382
def get_raw_records(self, memos_for_retrieval):
2383
"""Get the raw bytes for a records.
2385
:param memos_for_retrieval: An iterable containing the (index, pos,
2386
length) memo for retrieving the bytes. The Pack access method
2387
looks up the pack to use for a given record in its index_to_pack
2389
:return: An iterator over the bytes of the records.
2391
# first pass, group into same-index requests
2393
current_index = None
2394
for (index, offset, length) in memos_for_retrieval:
2395
if current_index == index:
2396
current_list.append((offset, length))
2398
if current_index is not None:
2399
request_lists.append((current_index, current_list))
2400
current_index = index
2401
current_list = [(offset, length)]
2402
# handle the last entry
2403
if current_index is not None:
2404
request_lists.append((current_index, current_list))
2405
for index, offsets in request_lists:
2406
transport, path = self._indices[index]
2407
reader = pack.make_readv_reader(transport, path, offsets)
2408
for names, read_func in reader.iter_records():
2409
yield read_func(None)
2411
def set_writer(self, writer, index, transport_packname):
2412
"""Set a writer to use for adding data."""
2413
if index is not None:
2414
self._indices[index] = transport_packname
2415
self._container_writer = writer
2416
self._write_index = index
2419
# Deprecated, use PatienceSequenceMatcher instead
2420
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2423
def annotate_knit(knit, revision_id):
2424
"""Annotate a knit with no cached annotations.
2426
This implementation is for knits with no cached annotations.
2427
It will work for knits with cached annotations, but this is not
2430
annotator = _KnitAnnotator(knit)
2431
return iter(annotator.annotate(revision_id))
2434
class _KnitAnnotator(object):
2435
"""Build up the annotations for a text."""
2437
def __init__(self, knit):
2440
# Content objects, differs from fulltexts because of how final newlines
2441
# are treated by knits. the content objects here will always have a
2443
self._fulltext_contents = {}
2445
# Annotated lines of specific revisions
2446
self._annotated_lines = {}
2448
# Track the raw data for nodes that we could not process yet.
2449
# This maps the revision_id of the base to a list of children that will
2450
# annotated from it.
2451
self._pending_children = {}
2453
# Nodes which cannot be extracted
2454
self._ghosts = set()
2456
# Track how many children this node has, so we know if we need to keep
2458
self._annotate_children = {}
2459
self._compression_children = {}
2461
self._all_build_details = {}
2462
# The children => parent revision_id graph
2463
self._revision_id_graph = {}
2465
self._heads_provider = None
2467
self._nodes_to_keep_annotations = set()
2468
self._generations_until_keep = 100
2470
def set_generations_until_keep(self, value):
2471
"""Set the number of generations before caching a node.
2473
Setting this to -1 will cache every merge node, setting this higher
2474
will cache fewer nodes.
2476
self._generations_until_keep = value
2478
def _add_fulltext_content(self, revision_id, content_obj):
2479
self._fulltext_contents[revision_id] = content_obj
2480
# TODO: jam 20080305 It might be good to check the sha1digest here
2481
return content_obj.text()
2483
def _check_parents(self, child, nodes_to_annotate):
2484
"""Check if all parents have been processed.
2486
:param child: A tuple of (rev_id, parents, raw_content)
2487
:param nodes_to_annotate: If child is ready, add it to
2488
nodes_to_annotate, otherwise put it back in self._pending_children
2490
for parent_id in child[1]:
2491
if (parent_id not in self._annotated_lines):
2492
# This parent is present, but another parent is missing
2493
self._pending_children.setdefault(parent_id,
2497
# This one is ready to be processed
2498
nodes_to_annotate.append(child)
2500
def _add_annotation(self, revision_id, fulltext, parent_ids,
2501
left_matching_blocks=None):
2502
"""Add an annotation entry.
2504
All parents should already have been annotated.
2505
:return: A list of children that now have their parents satisfied.
2507
a = self._annotated_lines
2508
annotated_parent_lines = [a[p] for p in parent_ids]
2509
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2510
fulltext, revision_id, left_matching_blocks,
2511
heads_provider=self._get_heads_provider()))
2512
self._annotated_lines[revision_id] = annotated_lines
2513
for p in parent_ids:
2514
ann_children = self._annotate_children[p]
2515
ann_children.remove(revision_id)
2516
if (not ann_children
2517
and p not in self._nodes_to_keep_annotations):
2518
del self._annotated_lines[p]
2519
del self._all_build_details[p]
2520
if p in self._fulltext_contents:
2521
del self._fulltext_contents[p]
2522
# Now that we've added this one, see if there are any pending
2523
# deltas to be done, certainly this parent is finished
2524
nodes_to_annotate = []
2525
for child in self._pending_children.pop(revision_id, []):
2526
self._check_parents(child, nodes_to_annotate)
2527
return nodes_to_annotate
2529
def _get_build_graph(self, key):
2530
"""Get the graphs for building texts and annotations.
2532
The data you need for creating a full text may be different than the
2533
data you need to annotate that text. (At a minimum, you need both
2534
parents to create an annotation, but only need 1 parent to generate the
2537
:return: A list of (key, index_memo) records, suitable for
2538
passing to read_records_iter to start reading in the raw data fro/
2541
if key in self._annotated_lines:
2544
pending = set([key])
2549
# get all pending nodes
2551
this_iteration = pending
2552
build_details = self._knit._index.get_build_details(this_iteration)
2553
self._all_build_details.update(build_details)
2554
# new_nodes = self._knit._index._get_entries(this_iteration)
2556
for key, details in build_details.iteritems():
2557
(index_memo, compression_parent, parents,
2558
record_details) = details
2559
self._revision_id_graph[key] = parents
2560
records.append((key, index_memo))
2561
# Do we actually need to check _annotated_lines?
2562
pending.update(p for p in parents
2563
if p not in self._all_build_details)
2564
if compression_parent:
2565
self._compression_children.setdefault(compression_parent,
2568
for parent in parents:
2569
self._annotate_children.setdefault(parent,
2571
num_gens = generation - kept_generation
2572
if ((num_gens >= self._generations_until_keep)
2573
and len(parents) > 1):
2574
kept_generation = generation
2575
self._nodes_to_keep_annotations.add(key)
2577
missing_versions = this_iteration.difference(build_details.keys())
2578
self._ghosts.update(missing_versions)
2579
for missing_version in missing_versions:
2580
# add a key, no parents
2581
self._revision_id_graph[missing_version] = ()
2582
pending.discard(missing_version) # don't look for it
2583
if self._ghosts.intersection(self._compression_children):
2585
"We cannot have nodes which have a ghost compression parent:\n"
2587
"compression children: %r"
2588
% (self._ghosts, self._compression_children))
2589
# Cleanout anything that depends on a ghost so that we don't wait for
2590
# the ghost to show up
2591
for node in self._ghosts:
2592
if node in self._annotate_children:
2593
# We won't be building this node
2594
del self._annotate_children[node]
2595
# Generally we will want to read the records in reverse order, because
2596
# we find the parent nodes after the children
2600
def _annotate_records(self, records):
2601
"""Build the annotations for the listed records."""
2602
# We iterate in the order read, rather than a strict order requested
2603
# However, process what we can, and put off to the side things that
2604
# still need parents, cleaning them up when those parents are
2606
for (rev_id, record,
2607
digest) in self._knit._read_records_iter(records):
2608
if rev_id in self._annotated_lines:
2610
parent_ids = self._revision_id_graph[rev_id]
2611
parent_ids = [p for p in parent_ids if p not in self._ghosts]
2612
details = self._all_build_details[rev_id]
2613
(index_memo, compression_parent, parents,
2614
record_details) = details
2615
nodes_to_annotate = []
2616
# TODO: Remove the punning between compression parents, and
2617
# parent_ids, we should be able to do this without assuming
2619
if len(parent_ids) == 0:
2620
# There are no parents for this node, so just add it
2621
# TODO: This probably needs to be decoupled
2622
fulltext_content, delta = self._knit._factory.parse_record(
2623
rev_id, record, record_details, None)
2624
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
2625
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
2626
parent_ids, left_matching_blocks=None))
2628
child = (rev_id, parent_ids, record)
2629
# Check if all the parents are present
2630
self._check_parents(child, nodes_to_annotate)
2631
while nodes_to_annotate:
2632
# Should we use a queue here instead of a stack?
2633
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
2634
(index_memo, compression_parent, parents,
2635
record_details) = self._all_build_details[rev_id]
2636
if compression_parent is not None:
2637
comp_children = self._compression_children[compression_parent]
2638
if rev_id not in comp_children:
2639
raise AssertionError("%r not in compression children %r"
2640
% (rev_id, comp_children))
2641
# If there is only 1 child, it is safe to reuse this
2643
reuse_content = (len(comp_children) == 1
2644
and compression_parent not in
2645
self._nodes_to_keep_annotations)
2647
# Remove it from the cache since it will be changing
2648
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
2649
# Make sure to copy the fulltext since it might be
2651
parent_fulltext = list(parent_fulltext_content.text())
2653
parent_fulltext_content = self._fulltext_contents[compression_parent]
2654
parent_fulltext = parent_fulltext_content.text()
2655
comp_children.remove(rev_id)
2656
fulltext_content, delta = self._knit._factory.parse_record(
2657
rev_id, record, record_details,
2658
parent_fulltext_content,
2659
copy_base_content=(not reuse_content))
2660
fulltext = self._add_fulltext_content(rev_id,
2662
blocks = KnitContent.get_line_delta_blocks(delta,
2663
parent_fulltext, fulltext)
2665
fulltext_content = self._knit._factory.parse_fulltext(
2667
fulltext = self._add_fulltext_content(rev_id,
2670
nodes_to_annotate.extend(
2671
self._add_annotation(rev_id, fulltext, parent_ids,
2672
left_matching_blocks=blocks))
2674
def _get_heads_provider(self):
2675
"""Create a heads provider for resolving ancestry issues."""
2676
if self._heads_provider is not None:
2677
return self._heads_provider
2678
parent_provider = _mod_graph.DictParentsProvider(
2679
self._revision_id_graph)
2680
graph_obj = _mod_graph.Graph(parent_provider)
2681
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
2682
self._heads_provider = head_cache
2685
def annotate(self, key):
2686
"""Return the annotated fulltext at the given key.
2688
:param key: The key to annotate.
2690
if True or len(self._knit._fallback_vfs) > 0:
2691
# stacked knits can't use the fast path at present.
2692
return self._simple_annotate(key)
2693
records = self._get_build_graph(key)
2694
if key in self._ghosts:
2695
raise errors.RevisionNotPresent(key, self._knit)
2696
self._annotate_records(records)
2697
return self._annotated_lines[key]
2699
def _simple_annotate(self, key):
2700
"""Return annotated fulltext, rediffing from the full texts.
2702
This is slow but makes no assumptions about the repository
2703
being able to produce line deltas.
2705
# TODO: this code generates a parent maps of present ancestors; it
2706
# could be split out into a separate method, and probably should use
2707
# iter_ancestry instead. -- mbp and robertc 20080704
2708
graph = _mod_graph.Graph(self._knit)
2709
head_cache = _mod_graph.FrozenHeadsCache(graph)
2710
search = graph._make_breadth_first_searcher([key])
2714
present, ghosts = search.next_with_ghosts()
2715
except StopIteration:
2717
keys.update(present)
2718
parent_map = self._knit.get_parent_map(keys)
2720
reannotate = annotate.reannotate
2721
for record in self._knit.get_record_stream(keys, 'topological', True):
2723
fulltext = split_lines(record.get_bytes_as('fulltext'))
2724
parents = parent_map[key]
2725
if parents is not None:
2726
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
2729
parent_cache[key] = list(
2730
reannotate(parent_lines, fulltext, key, None, head_cache))
2732
return parent_cache[key]
2734
raise errors.RevisionNotPresent(key, self._knit)
2738
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2740
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
